kernel/ucount.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2
   3 #include <linux/stat.h>
   4 #include <linux/sysctl.h>
   5 #include <linux/slab.h>
   6 #include <linux/cred.h>
   7 #include <linux/hash.h>
   8 #include <linux/kmemleak.h>
   9 #include <linux/user_namespace.h>
  10
  11 struct ucounts init_ucounts = {
  12         .ns    = &init_user_ns,
  13         .uid   = GLOBAL_ROOT_UID,
  14         .count = ATOMIC_INIT(1),
  15 };
  16
  17 #define UCOUNTS_HASHTABLE_BITS 10
  18 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
  19 static DEFINE_SPINLOCK(ucounts_lock);
  20
  21 #define ucounts_hashfn(ns, uid)                                         \
  22         hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
  23                   UCOUNTS_HASHTABLE_BITS)
  24 #define ucounts_hashentry(ns, uid)      \
  25         (ucounts_hashtable + ucounts_hashfn(ns, uid))
  26
  27
  28 #ifdef CONFIG_SYSCTL
  29 static struct ctl_table_set *
  30 set_lookup(struct ctl_table_root *root)
  31 {
  32         return &current_user_ns()->set;
  33 }
  34
  35 static int set_is_seen(struct ctl_table_set *set)
  36 {
  37         return &current_user_ns()->set == set;
  38 }
  39
  40 static int set_permissions(struct ctl_table_header *head,
  41                                   struct ctl_table *table)
  42 {
  43         struct user_namespace *user_ns =
  44                 container_of(head->set, struct user_namespace, set);
  45         int mode;
  46
  47         /* Allow users with CAP_SYS_RESOURCE unrestrained access */
  48         if (ns_capable(user_ns, CAP_SYS_RESOURCE))
  49                 mode = (table->mode & S_IRWXU) >> 6;
  50         else
  51         /* Allow all others at most read-only access */
  52                 mode = table->mode & S_IROTH;
  53         return (mode << 6) | (mode << 3) | mode;
  54 }
  55
  56 static struct ctl_table_root set_root = {
  57         .lookup = set_lookup,
  58         .permissions = set_permissions,
  59 };
  60
  61 #define UCOUNT_ENTRY(name)                              \
  62         {                                               \
  63                 .procname       = name,                 \
  64                 .maxlen         = sizeof(int),          \
  65                 .mode           = 0644,                 \
  66                 .proc_handler   = proc_dointvec_minmax, \
  67                 .extra1         = SYSCTL_ZERO,          \
  68                 .extra2         = SYSCTL_INT_MAX,       \
  69         }
  70 static struct ctl_table user_table[] = {
  71         UCOUNT_ENTRY("max_user_namespaces"),
  72         UCOUNT_ENTRY("max_pid_namespaces"),
  73         UCOUNT_ENTRY("max_uts_namespaces"),
  74         UCOUNT_ENTRY("max_ipc_namespaces"),
  75         UCOUNT_ENTRY("max_net_namespaces"),
  76         UCOUNT_ENTRY("max_mnt_namespaces"),
  77         UCOUNT_ENTRY("max_cgroup_namespaces"),
  78         UCOUNT_ENTRY("max_time_namespaces"),
  79 #ifdef CONFIG_INOTIFY_USER
  80         UCOUNT_ENTRY("max_inotify_instances"),
  81         UCOUNT_ENTRY("max_inotify_watches"),
  82 #endif
  83 #ifdef CONFIG_FANOTIFY
  84         UCOUNT_ENTRY("max_fanotify_groups"),
  85         UCOUNT_ENTRY("max_fanotify_marks"),
  86 #endif
  87         { },
  88         { },
  89         { },
  90         { },
  91         { }
  92 };
  93 #endif /* CONFIG_SYSCTL */
  94
  95 bool setup_userns_sysctls(struct user_namespace *ns)
  96 {
  97 #ifdef CONFIG_SYSCTL
  98         struct ctl_table *tbl;
  99
 100         BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
 101         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
 102         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
 103         if (tbl) {
 104                 int i;
 105                 for (i = 0; i < UCOUNT_COUNTS; i++) {
 106                         tbl[i].data = &ns->ucount_max[i];
 107                 }
 108                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
 109         }
 110         if (!ns->sysctls) {
 111                 kfree(tbl);
 112                 retire_sysctl_set(&ns->set);
 113                 return false;
 114         }
 115 #endif
 116         return true;
 117 }
 118
 119 void retire_userns_sysctls(struct user_namespace *ns)
 120 {
 121 #ifdef CONFIG_SYSCTL
 122         struct ctl_table *tbl;
 123
 124         tbl = ns->sysctls->ctl_table_arg;
 125         unregister_sysctl_table(ns->sysctls);
 126         retire_sysctl_set(&ns->set);
 127         kfree(tbl);
 128 #endif
 129 }
 130
 131 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
 132 {
 133         struct ucounts *ucounts;
 134
 135         hlist_for_each_entry(ucounts, hashent, node) {
 136                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
 137                         return ucounts;
 138         }
 139         return NULL;
 140 }
 141
 142 static void hlist_add_ucounts(struct ucounts *ucounts)
 143 {
 144         struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
 145         spin_lock_irq(&ucounts_lock);
 146         hlist_add_head(&ucounts->node, hashent);
 147         spin_unlock_irq(&ucounts_lock);
 148 }
 149
 150 struct ucounts *get_ucounts(struct ucounts *ucounts)
 151 {
 152         if (ucounts && atomic_add_negative(1, &ucounts->count)) {
 153                 put_ucounts(ucounts);
 154                 ucounts = NULL;
 155         }
 156         return ucounts;
 157 }
 158
 159 struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
 160 {
 161         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
 162         struct ucounts *ucounts, *new;
 163
 164         spin_lock_irq(&ucounts_lock);
 165         ucounts = find_ucounts(ns, uid, hashent);
 166         if (!ucounts) {
 167                 spin_unlock_irq(&ucounts_lock);
 168
 169                 new = kzalloc(sizeof(*new), GFP_KERNEL);
 170                 if (!new)
 171                         return NULL;
 172
 173                 new->ns = ns;
 174                 new->uid = uid;
 175                 atomic_set(&new->count, 1);
 176
 177                 spin_lock_irq(&ucounts_lock);
 178                 ucounts = find_ucounts(ns, uid, hashent);
 179                 if (ucounts) {
 180                         kfree(new);
 181                 } else {
 182                         hlist_add_head(&new->node, hashent);
 183                         spin_unlock_irq(&ucounts_lock);
 184                         return new;
 185                 }
 186         }
 187         spin_unlock_irq(&ucounts_lock);
 188         ucounts = get_ucounts(ucounts);
 189         return ucounts;
 190 }
 191
 192 void put_ucounts(struct ucounts *ucounts)
 193 {
 194         unsigned long flags;
 195
 196         if (atomic_dec_and_test(&ucounts->count)) {
 197                 spin_lock_irqsave(&ucounts_lock, flags);
 198                 hlist_del_init(&ucounts->node);
 199                 spin_unlock_irqrestore(&ucounts_lock, flags);
 200                 kfree(ucounts);
 201         }
 202 }
 203
 204 static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
 205 {
 206         long c, old;
 207         c = atomic_long_read(v);
 208         for (;;) {
 209                 if (unlikely(c >= u))
 210                         return false;
 211                 old = atomic_long_cmpxchg(v, c, c+1);
 212                 if (likely(old == c))
 213                         return true;
 214                 c = old;
 215         }
 216 }
 217
 218 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
 219                            enum ucount_type type)
 220 {
 221         struct ucounts *ucounts, *iter, *bad;
 222         struct user_namespace *tns;
 223         ucounts = alloc_ucounts(ns, uid);
 224         for (iter = ucounts; iter; iter = tns->ucounts) {
 225                 long max;
 226                 tns = iter->ns;
 227                 max = READ_ONCE(tns->ucount_max[type]);
 228                 if (!atomic_long_inc_below(&iter->ucount[type], max))
 229                         goto fail;
 230         }
 231         return ucounts;
 232 fail:
 233         bad = iter;
 234         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
 235                 atomic_long_dec(&iter->ucount[type]);
 236
 237         put_ucounts(ucounts);
 238         return NULL;
 239 }
 240
 241 void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
 242 {
 243         struct ucounts *iter;
 244         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 245                 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
 246                 WARN_ON_ONCE(dec < 0);
 247         }
 248         put_ucounts(ucounts);
 249 }
 250
 251 long inc_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 252 {
 253         struct ucounts *iter;
 254         long ret = 0;
 255
 256         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 257                 long max = READ_ONCE(iter->ns->ucount_max[type]);
 258                 long new = atomic_long_add_return(v, &iter->ucount[type]);
 259                 if (new < 0 || new > max)
 260                         ret = LONG_MAX;
 261                 else if (iter == ucounts)
 262                         ret = new;
 263         }
 264         return ret;
 265 }
 266
 267 bool dec_rlimit_ucounts(struct ucounts *ucounts, enum ucount_type type, long v)
 268 {
 269         struct ucounts *iter;
 270         long new = -1; /* Silence compiler warning */
 271         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 272                 long dec = atomic_long_add_return(-v, &iter->ucount[type]);
 273                 WARN_ON_ONCE(dec < 0);
 274                 if (iter == ucounts)
 275                         new = dec;
 276         }
 277         return (new == 0);
 278 }
 279
 280 bool is_ucounts_overlimit(struct ucounts *ucounts, enum ucount_type type, unsigned long max)
 281 {
 282         struct ucounts *iter;
 283         if (get_ucounts_value(ucounts, type) > max)
 284                 return true;
 285         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
 286                 max = READ_ONCE(iter->ns->ucount_max[type]);
 287                 if (get_ucounts_value(iter, type) > max)
 288                         return true;
 289         }
 290         return false;
 291 }
 292
 293 static __init int user_namespace_sysctl_init(void)
 294 {
 295 #ifdef CONFIG_SYSCTL
 296         static struct ctl_table_header *user_header;
 297         static struct ctl_table empty[1];
 298         /*
 299          * It is necessary to register the user directory in the
 300          * default set so that registrations in the child sets work
 301          * properly.
 302          */
 303         user_header = register_sysctl("user", empty);
 304         kmemleak_ignore(user_header);
 305         BUG_ON(!user_header);
 306         BUG_ON(!setup_userns_sysctls(&init_user_ns));
 307 #endif
 308         hlist_add_ucounts(&init_ucounts);
 309         inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
 310         return 0;
 311 }
 312 subsys_initcall(user_namespace_sysctl_init);